Multi-core processor-based multivariate deep network model reconstruction method and device

Abstract

The invention discloses a multi-core processor-based multivariate deep network model reconstruction method and device. The method includes: acquiring a video stream image collected by a camera; selecting a logical combination relationship, and determining the cascading relationship between each deep network model in the deep network module and the corresponding output action according to the logical combination relationship; loading the corresponding deep network model; Call the multi-core dynamic allocation management command to calculate the complexity of the loaded deep network model, allocate corresponding memory and predetermined number of core processors for each deep network model according to the complexity; input the collected video stream image into the corresponding deep network model ; Analyze the scene information in the video stream image according to the specified output information obtained after being processed by each cascaded deep network model, and execute the corresponding output action. The invention requires low memory space, flexibly combines various deep network models to form a building block development mode, and improves user development efficiency and interest.

Description

technical field [0001] The invention relates to data processing technology, in particular to data processing under a deep network model using a multi-core processor, and in particular to a method and device for reconstructing a multi-component deep network model based on a multi-core processor. Background technique [0002] Existing deep network models often can only handle tasks such as detection, classification, and segmentation of a single scene. It will be powerless when it comes to complex scenes. [0003] There are usually two ways to solve complex scene analysis in the existing methods: one is to use a relatively high-cost GPU to directly output the output information of multiple deep network models in series. However, the high cost of GPU is not conducive to large-scale use; the other is to directly analyze the timing of the scene, and analyze the scene through the spatial position correlation and temporal correlation between actions or objects in the scene. The tr...

AI Technical Summary

Problems solved by technology

However, the cost of GPU is too high, which is not conducive to large-scale use; the other is to directly analyze the timing of the scene, and analyze the scene through the spatial position correlation and time correlation between actions or objects in the scene.

The training data set required by this method is relatively special, which is cumbersome to process, and the continuous timing analysis of the video requires a large memory space and high algorithm complexity.

Images